Παρακολούθηση κατεργασιών τόρνευσης με τεχνικές μηχανικής μάθησης

Καψής, Χαράλαμπος; Kapsis, Charalampos

dc.contributor.author	Καψής, Χαράλαμπος	el
dc.contributor.author	Kapsis, Charalampos	en
dc.date.accessioned	2026-02-18T12:54:07Z
dc.date.available	2026-02-18T12:54:07Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/63504
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.31199
dc.rights	Default License
dc.subject	Mechanical Engineering	en
dc.subject	Turning Process	en
dc.subject	Artificial Intelligence	en
dc.subject	Data Processing	en
dc.subject	Industry 4.0	en
dc.subject	Μηχανολογία	el
dc.subject	Τεχνητή Νοημοσύνη	el
dc.subject	Τόρνευση	el
dc.subject	Επεξεργασία Δεδομένων	el
dc.subject	4η Βιομηχανική Επανάσταση	el
dc.title	Παρακολούθηση κατεργασιών τόρνευσης με τεχνικές μηχανικής μάθησης	el
heal.type	masterThesis
heal.secondaryTitle	Turning process monitoring using machine learning techniques	el
heal.generalDescription	Στο Εργαστήριο Τεχνολογίας των Κατεργασιών έχουν γίνει πρόσφατα μετρήσεις ποιότητας τεμαχίων που κατασκευάστηκαν με κατεργασία τόρνευσης και συσχετισμός τους με τις παραμέτρους προγραμματισμού της κατεργασίας καθώς και με μετρήσεις αισθητήρων επιτάχυνσης, ροπής κινητήρων κλπ. με βάση τις οποίες παρακολουθείται η κατάσταση της κατεργασίας. Στη συνέχεια κατασκευάστηκαν συνελικτικά νευρωνικά δίκτυα που προέβλεπαν την ποιότητα του τεμαχίου από τις καταγεγραμμένες μετρήσεις των αισθητήρων (raw data – ανεπεξέργαστα δεδομένα) χωρίς ενδιάμεση αναγνώριση μορφολογικών χαρακτηριστικών (features).	el
heal.classification	Mechanical Engineering & Artificial Intelligence	en
heal.classification	Μηχανολογία και Τεχνητή Νοημοσύνη	el
heal.language	el
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2025-06
heal.abstract	The Fourth Industrial Revolution (Industry 4.0) has highlighted the need to integrate intelligent technologies into the production process, aiming at automation, performance optimization, and enhancement of product quality. Within this context, the present work focuses on the development and evaluation of an intelligent monitoring system for turning operations, based on machine learning techniques and signal analysis, with the goal of automatically classifying the quality of the machined parts. The adopted methodology includes experimental machining on a CNC lathe, during which signals were collected from the machine tool controller (speed and torque along the X, Z axes and the spindle), as well as from accelerometers mounted on the same axes. Signal acquisition was carried out through a DAQ system with a sampling rate of 1 kHz, ensuring accuracy and synchronization. The signals underwent preliminary processing, including denoising using wavelet analysis (Wavelet Denoising – Coiflet2), and feature extraction from both the time domain (RMS, Energy, Kurtosis, Skewness, ZCR, Number of Peaks, Envelope RMS, Envelope Energy, Entropy) and the frequency domain (Dominant Frequency, Spectral Centroid, Total Spectrum Power). The extracted features were used as input data for training machine learning algorithms. The quality of the machined parts was categorized into three classes (Low, Medium, High), based on metrological parameters such as surface roughness (Ra, Rz), dimensional accuracy, and cutting depth. Among these, the main target variable (label) selected was the Roughness_Ra_Class. Two core classification models were implemented: the Random Forest Classifier and a Multi-Layer Perceptron (MLP). Multiple implementations were carried out: (a) using the initial imbalanced dataset, (b) applying manual balancing (undersampling/oversampling), and (c) using an enriched feature set. Feature selection techniques were applied to optimize performance. Several alternative configurations were also tested (different models, application of the SMOTE technique to address class imbalance, model tuning, etc.) in order to achieve better results. The results demonstrated that careful feature selection and extraction, combined with appropriate preprocessing and dataset balancing, significantly improve the models’ ability to accurately predict the quality of the machined parts. In conclusion, this study confirms that utilizing vibration, speed, and torque signals from the lathe through machine learning techniques enables effective estimation of the quality of machined components. This approach can be integrated into modern industrial production lines, offering enhanced quality control, waste reduction, and increased equipment autonomy — fully aligned with the principles of Industry 4.0.	el
heal.advisorName	Βοσνιάκος, Χριστόφορος-Γεώργιος	el
heal.committeeMemberName	Μπενάρδος, Πανώριος	el
heal.committeeMemberName	Κουλουριώτης, Δημήτριος	el
heal.committeeMemberName	Βοσνιάκος, Χριστόφορος-Γεώργιος	el
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Μηχανολόγων Μηχανικών	el
heal.academicPublisherID	ntua
heal.numberOfPages	56 σ.	el
heal.fullTextAvailability	false